The Home Office wants to use AI facial age estimation in some disputed asylum age decisions from 2027. More than 100 refugee and children’s organisations have warned that the plan could increase the risk of children being wrongly treated as adults.
For ordinary UK readers, this is an important AI story because it is not about a chatbot writing emails or a phone adding clever photo tricks. It is about using a machine estimate in a high-stakes public decision: whether someone should be treated as a child, with child protections, or as an adult.
The technology is called facial age estimation, or FAE. The basic idea is simple enough. A system looks at a photo of a person’s face and estimates their age. The Home Office says this would be used only as extra information for immigration officers, not as an automatic decision. Campaigners argue that even an “extra” signal can carry too much weight when a decision is already difficult and urgent.
That tension is where the real lesson sits. AI does not have to be fully in charge to affect people’s lives. Sometimes the bigger question is how much confidence a computer-generated number creates in the human who still has to make the call.
What the Home Office says it plans to do
The government’s new guidance, published on 29 May 2026, says facial age estimation is not yet in operational use. The Home Office says it is testing the technology during 2026, with a view to using it at the border in 2027.
The stated use is narrow: disputed age cases where someone arriving in the UK says they are under 18 and there is no reliable document proving their age. The Home Office says age decisions can be hard because people may arrive without papers, may not know their exact date of birth, or may look older or younger than they are.
At the moment, immigration officers can make an initial age decision using appearance and demeanour. The Home Office says officers apply a benefit-of-the-doubt approach: a person should only be treated as an adult at this first stage if two officers independently think their physical appearance and demeanour very strongly suggest they are significantly over 18.
The government says FAE would add another piece of information. It also stresses that the technology is not facial recognition. Facial recognition tries to identify who someone is by comparing faces. Facial age estimation tries to estimate how old someone is from an image.
That difference matters, but it does not remove the difficult question. Estimating age is still a judgement about a person, and the consequences of getting it wrong can be serious.
Why charities are worried
The Guardian reported on 1 June 2026 that a coalition of more than 100 organisations had criticised the plan. Their concern is that young people could be placed in adult detention or other adult settings if the technology, or the way it is used, pushes an uncertain case in the wrong direction.
That risk is not abstract. The age boundary around 16, 17, 18 and 19 is exactly where small errors matter most. A technology that is broadly useful for checking whether someone looks safely over 25 at a supermarket till is being asked a more sensitive question when the answer may affect a young asylum seeker’s protection, housing, care and legal process.
The Home Office guidance itself acknowledges limits. It says there is no single method that can determine a person’s exact age without official documents. It also notes that accuracy can be affected by image quality, age group, gender and where someone comes from, and that even leading systems have a margin of error near the 16-to-18 boundary.
That is why this story is not simply “AI good” or “AI bad”. The practical issue is governance: who uses the estimate, how it is explained, what happens when the machine and the human disagree, and how a person can challenge a decision that may partly rest on a system they cannot inspect.
Why this matters beyond asylum
Most readers will never see this exact process. But many will encounter facial age estimation elsewhere. The same broad family of tools is already used or trialled for age-restricted websites, social media accounts and age-restricted purchases.
That is why the asylum plan is a useful case study in how everyday AI systems move from convenience into authority. A face scan at an online checkout may feel like a quick alternative to showing ID. A face scan at a border is very different, because the person being scanned may have little power, little time, and a lot at stake.
This is a pattern ManyHands has covered before with AI tools that get more access and authority. The more important the task, the more we need to ask what the system can actually do, who supervises it, and what happens when it is wrong.
It also connects to the wider problem of trusting automated judgements too quickly. When an AI system gives a neat answer, it can make uncertainty look tidier than it really is. That can be helpful in a low-risk setting, but risky in a safeguarding decision.
What ordinary readers should ask
Is the AI deciding, or advising? The Home Office says officers will still decide. That is important, but readers should still ask how much weight the AI estimate will carry in practice. A tool can shape a decision even when it is formally described as supportive.
How accurate is it for the people affected? A general accuracy claim is not enough. Age estimation near 18 is harder than broad age checks, and performance can vary by image quality and demographics. The key question is how the system performs in the exact setting where it will be used.
What is the appeal route? If a person is treated as an adult after a disputed age decision, there needs to be a clear way to challenge that outcome. That includes understanding whether an AI estimate was used and how it influenced the decision.
What happens to the image? The government says facial age estimation is different from facial recognition and does not identify a person by searching a database. Readers should still expect clear answers about photo capture, storage, deletion, audit logs and who can access the data.
Who checks the checker? Independent testing matters. So does ongoing monitoring once a tool is used in real cases. A system can look acceptable in a controlled test and still behave differently in busy, messy public-service conditions.
A human decision can still be changed by a machine
One phrase to watch in AI policy is “human in the loop”. It sounds reassuring, and sometimes it is. But a human decision-maker may be tired, under time pressure, short of information or encouraged by an official tool to feel more certain than the evidence deserves.
That does not mean public bodies should never use AI. It means the phrase “a human remains responsible” is the beginning of the conversation, not the end. The public needs to know what the human sees, how the AI result is framed, and whether there is space to say: the machine estimate is not strong enough here.
This is especially important when children may be involved. Safeguarding decisions often require caution precisely because certainty is hard. If AI is introduced, it should make that caution easier to apply, not harder.
The practical takeaway
The Home Office’s plan shows how AI is moving into decisions where a wrong answer can have life-changing consequences. The technology may be presented as a quick, repeatable aid, but the real test is whether it improves fairness and protection in practice.
For UK readers, the useful habit is to look past the word “AI” and ask plainer questions. What decision is being made? Who is affected? How wrong can the system be? Can a person challenge it? Is the technology being used because it is genuinely safer, or because it is faster and cheaper?
Those questions matter whether the AI is estimating age at a border, summarising a medical query, moderating a social media account or helping with a workplace decision. The higher the stakes, the less impressed we should be by speed alone.
Sources: The Guardian and GOV.UK.
